Development and application of a new clustering-based archive reduction method for the design of multi-objective optimization algorithms

The main challenge in solving multi-objective optimization problems with conflicting objective functions is to find the global optimal solution set in a stable and efficient way. Studies in the literature report that the crowd distance method fails to provide diversity in the search space. In this thesis, a new archive reduction method is proposed to improve the search performance of Pareto-based multi-objective evolutionary search algorithms on multimodal multi-objective optimization problems. In the proposed method, the archive reduction process is performed by a clustering algorithm based on dynamic switched reference spaces. In this process, the task of the clustering mechanism is to group the vectors according to their similarity and update the archive by selecting only one vector from each cluster when the number of vectors in the archive is exceeded. By dynamically determining the reference space vector for the clustering process, the decision and objective spaces can be varied. According to the results obtained from the test runs, the algorithm designed has better performance metric values than both the base algorithm, its improved version and its strong competitors.